dirty little secrets
DESCRIPTION
Dirty Little SecretsTRANSCRIPT
-
DEPARTMENT OF ECONOMICS OxCarre Oxford Centre for the Analysis of Resource Rich Economies Manor Road Building, Manor Road, Oxford OX1 3UQ Tel: +44(0)1865 281281 Fax: +44(0)1865 271094 [email protected] www.oxcarre.ox.ac.uk
Direct tel: +44(0) 1865 281281 E-mail: [email protected]
_
OxCarre Research Paper 134
Dirty Little Secrets: Inferring Fossil-Fuel Subsidies from Patterns in Emission
Intensities
Radoslaw (Radek) Stefanski* Laval University
*OxCarre External Research Associate
-
Dirty Little Secrets: Inferring Fossil-Fuel Subsidies from
Patterns in Emission Intensities1
Radoslaw (Radek) Stefanski
Laval University and University of Oxford (OxCarre)
April 4, 2014
Abstract
I develop a unique database of international fossil-fuel subsidies by examining country-
specific patterns in carbon emission-to-GDP ratios, known as emission-intensities. For most
but not all countries, intensities tend to be hump-shaped with income. I construct a model
of structural-transformation that generates this hump-shaped intensity and then show that
deviations from this pattern must be driven by distortions to sectoral-productivity and/or
fossil-fuel prices. Finally, I use the calibrated model to measure these distortions for 170
countries for 1980-2010. This methodology reveals that fossil-fuel price-distortions are large,
increasing and often hidden. Furthermore, they are major contributors to higher carbon-
emissions and lower GDP.
1 An earlier version of this paper was circulated under the title Structural Transformation and Pollution.I would like to thank Thierry Brechet, Mario Crucini, Marine Ganofsky, Torfinn Harding, Philipp Kircher, DavidLagakos, Peter Neary, Bill Nordhaus, Fabrizio Perri, Rick van der Ploeg, Tony Smith and Tony Venables as wellas seminar participants at the University of Minnesota, the University of Oxford, Yale University, Universityof Calgary, Edinburgh University, University of St Andrews, Pontificia Universidad Catolica de Chile, LavalUniversity, Universite de Sherbrooke and the University of Surrey for helpful comments and discussion. I havealso benefited from comments of seminar participants at the Overlapping Generations Days Meetings (Vielsalm),the World Congress of Environmental and Resource Economists (Montreal), the Royal Economic Society Meetings(Cambridge), NEUDC 2013 (Boston), Midwest Macro (Minneapolis) and the Canadian Economics AssociationMeetings (Montreal). All errors are my own. Contact: [email protected]
-
1 Introduction
An astonishing feature of international energy and climate policy is that fossil fuels - often seen
as the primary contributor to climate change - receive enormous government support.2 Elim-
inating these distortionary policies could in principle improve efficiency, provide a reprieve to
strained government budgets whilst also lowering carbon emissions.3 Surprisingly, no compre-
hensive database of directly measured, comparable fossil-fuel subsidies exists at the international
level. As argued by Koplow (2009), this is both because of political pressure from the direct
beneficiaries of subsidies and because of the immense complexity of the task given the profusion
and diversity of subsidy programs across countries.4 Indirect measures of subsidies - such as
the ones constructed by the IMF (2013) or the IEA (2012) - are based on the price-gap ap-
proach. This methodology allows researchers to infer national subsidies by comparing measured
energy prices with an international benchmark price. The key limitation of this technique is
that it does not account for government actions which support carbon energy without changing
its final price (Koplow, 2009).5 Furthermore, the data necessary for this exercise is limited and
since estimates are based on energy prices measured at the pump, they incorporate significant
non-traded components which biases estimates. In this paper, I develop a completely novel, in-
direct, model-based method for inferring these carbon fossil-fuel wedges. I do this by examining
country-specific patterns in carbon emission-to-GDP ratios, known as emission intensities.
The method is based on two observations about carbon emission intensity. First, emission
intensities follow a robust hump-shaped pattern with income. Figure 1(a) plots total CO2
emissions per dollar of GDP for 26 OECD countries versus each countrys GDP per capita, for
1751-2010. The graph suggests that middle-income countries produce dirtier output than rich
or poor countries. Second, the emission intensity of later developers tends to follow a so-called
envelope-pattern over time: the intensities of later developers rise quickly until they roughly
reach the intensity of the UK - the first country to start the modern development process -
after which, their intensity tends to approximately follow the same path as that of the UK. An
illustrative example of this envelope-pattern is shown in Figure 1(b). In the graph, the obvious
exceptions are China and the former USSR, which greatly overshoot this pattern.6 In this paper
I argue that the extent to which countries like China deviate from the hump-shaped pattern,
2 Rough, lower-bound estimates by IMF (2013) show that global fossil fuel subsidies in 2009 were on theorder of magnitude of US$ 480 billion.
3 See, for example, IEA (2012), OECD (2012), IMF (2013) or Koplow (2009).4 Work by OECD (2012) is the only attempt to directly calculate carbon subsidies. These estimates, however,
are only for a select number of countries and years and they are not comparable across countries.5 For example in the US, oil and gas producers receive support if they have older technology or access to
more expensive reserves. As argued by Koplow (2009), the subsidy is not likely to change the market price ofheating oil or gasoline, simply because the subsidized producer is a very small player in the global oil market.
6 Notice that whilst the above are illustrative, the hump-shape and envelope patterns are statistically robustas is shown in the Appendix.
-
31
10
100
1000
0 10000 20000 30000GDP per Capita (1990 GK US $)
T on s
of C
a rb o
n /M
i l li o
n P P
P U
S $
CO2 Emissions per unit of GDP
(a) OECD CO2 Emission Intensity, 1751-2010
10
100
1000
10000
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000To n
s o f
Ca r
b on /
Mi l l
i on
P PP
US $ CO2 Intensity
UK
US Canada KoreaChina
USSR
(b) Timing and Emission Intensity
Figure 1: Carbon Dioxide Emission Intensity Patterns
is indicative of different types of distortions within those economies. I then demonstrate how a
simple model can be used to measure these distortions.
To do this, I construct a model of structural transformation calibrated to the experience
of the UK. The model reproduces the hump-shaped emission intensity by generating an en-
dogenously changing fuel mix and energy intensity. I then examine cross-country differences in
emission intensity through the lens of the model. In my framework, any deviation in a countrys
emission intensity from the hump shape pattern is indicative of one of three distortions or wedges
within that economy: 1) a wedge to agricultural productivity, 2) a wedge to non-agricultural
productivity and a 3) subsidy-like wedge to fossil fuel prices. Following the language of Chari et
al. (2007) and Duarte and Restuccia (2007), these wedges are objects that appear like shocks
to productivity or prices in a standard model but in fact reflect a wider set of distortions,
imperfections or government policies found in the data.
The contribution of the paper is to show that the envelope pattern in CO2 emission intensities
is a consequence of different starting dates of industrialization, which in turn are driven by cross-
country wedges in agricultural productivity. Any other deviations from the hump-shaped pattern
are symptomatic of either non-agricultural productivity wedges or subsidy-like wedges on fossil
fuels. Given the calibrated, structural model I can then use data on a countrys CO2 intensity,
the size of its agricultural sector and its GDP levels to measure the size of these three wedges -
and in particular I can infer the size of the energy subsidy wedge across countries and over time.
Before I construct a model that generates a hump-shaped intensity and use it to extract
energy wedges, I first need to isolate the key drivers of emission intensity. To do this, I perform
an accounting exercise on a panel of international data. I show that the hump-shape emission
intensity is driven by two factors. First, the increasing part of the hump shape stems from a
-
4changing fuel mix. As countries grow richer, they tend to shift from using clean, carbon-neutral
bio-fuels (like wood) towards dirty, carbon positive fuels (like coal). This generates a rising
impurity of fuels and an increasing emission intensity. Second, I demonstrate that in the data
the declining part of emission intensity stems from falling energy intensity - the energy to GDP
ratio of an economy. This captures the idea that over time an economy needs less energy to
produce the same amount of output and hence will release less carbon per unit of output.
To reproduce the above two mechanisms which drive emission intensity, I build and cali-
brate a two-sector, general equilibrium growth model of structural transformation similar to
Gollin et al. (2002), Rogerson (2007) or Duarte and Restuccia (2007) but with energy as an
intermediate input. First, to capture changing fuel mix I introduce non-homothetic preferences
in agriculture and assume that agriculture uses only bio-fuels, whilst non-agriculture uses only
fossil fuels. Increasing agricultural productivity generates a structural transformation away from
agriculture towards non-agriculture which in turn drives a change in the fuel mix resulting in
increasing impurity and hence rising emission intensity. Second, I generate falling energy in-
tensity through substitution effects due to differential productivity growth rates across energy
and non-energy sectors. Unlike other inputs, energy is used at every stage of production. Since
inputs likely benefit from technological progress at each stage of production, energy will benefit
disproportionately more from technological progress than other inputs. If, additionally, energy
and non-energy inputs are complements in production, then higher technological progress asso-
ciated with energy will lead to a shift of resources away from energy towards non-energy and
result in falling energy intensity. Together, an increasingly dirty fuel mix and a falling energy
intensity can give rise to a hump-shaped emission intensity. I calibrate the model to match the
structural transformation and energy consumption patterns of the UK between 1820 and 2010.
Finally, I use the calibrated model to derive the quantitative role that wedges play in de-
termining emission intensity outcomes across countries. In this exercise, I maintain preference
and elasticity parameters from the baseline UK economy but I introduce three wedges to match
specific features of individual countries. In particular, I choose: 1) the agricultural productiv-
ity wedge to match the share of agricultural employment; 2) the non-agricultural productivity
wedge to match aggregate labor productivity; and 3) the energy subsidy wedge to match any
remaining differences in emission intensity across countries. I then examine the contribution of
each wedge to the deviation of a countrys emission intensity from the intensity of the United
Kingdom.
I find that the agricultural productivity wedge will delay the start of structural transforma-
tion. Countries moving out of agriculture later, will be in a world where the non-agricultural
sector is less energy intensive and hence has a lower emission intensity. Cross-country differences
in agricultural productivity will thus generate the envelope pattern in intensity observed in the
-
5data. The impact of the other two wedges on intensity depends on model-specific parameters,
but - in general - both wedges give rise to more substitution towards energy and hence higher
energy intensities. In this way, I use a calibrated model of structural transformation to disen-
tangle the different types of distortions found in an economy, and to infer the size of fossil fuel
energy wedges implied by the deviation from the UKs hump-shape pattern for a panel of 170
countries over the 1980-2010 period.
There are three key benefits to using the above method. First, the measure of energy
distortions calculated above will be a residual and hence will be wider than a direct measurement
of fossil fuel subsidies.7 This is helpful since it provides a broader picture of the extent of
support to fossil fuels around the world and allows us to infer subsidies that do not directly
affect energy prices. Second, the approach overcomes an important issue of data scarcity. The
standard price-gap method of inferring fossil fuel subsidies depends on knowing the cost of fossil
fuels at-the-pump. This data is limited to a few years, whilst my approach can easily provide
measures of fossil fuel wedges for many decades. Finally, and perhaps most importantly, the
model-based approach allows me to perform counterfactuals and to examine how both output
and emissions would have evolved without energy wedges in place.
Examining the fossil fuel wedges obtained from the model, I find that the size of subsidies is
enormous - 983 billion (1990 PPP) US dollars in 2010 alone. Wedges have also been growing -
more than quadrupling since the late 90s. Crucially, I find evidence of large, indirect subsidies
in some countries - like China. The model suggests that support to fossil fuel energy in those
countries is not reflected in prices of petrol at the pump, but is rather indirect. This matches
well with earlier studies - like Zhao (2001) - who find that the Chinese government supports
energy intensive industries through a wide range of indirect subsidies. Finally, I perform a
counterfactual were I turn off energy wedges in each country and find that up to 36% of global
carbon emissions between 1980 and 2010 were driven by subsidies and that GDP was up to 1.7%
lower per year because of the distortive subsidies.
In the following section, I perform a pollution accounting exercise demonstrating the impor-
tance of changing fuel mix and structural transformation to hump-shaped intensity. Sections 3
and 4 present the model and its solution. Section 5 calibrates the model and demonstrates that
the mechanism does well in accounting for the observed intensity of the UK. Section 6 presents
examples of how distortions in the original economy result in changes in emission intensity.
Section 7 uses these findings to measure subsidies for a wide panel of countries. Sections 8 ex-
amines the resulting subsidies, whilst section 9 considers the counterfactual case when subsidies
are eliminated. Finally section 10 performs robustness exercises and section 11 concludes.
7 In particular it will also capture indirect government support as well as differences in country-specific factorssuch as transportation costs, resource endowments, tariffs or other policies or geographical features causingcountries to have a disproportionately dirtier or cleaner output.
-
62 Stylized Facts
In this section I pinpoint the mechanisms driving the hump-shaped CO2 emission intensity
documented above by performing a pollution accounting exercise on UK data.8 The findings
of this section will be used to motivate the model constructed in the next section and in turn, the
model will be used to analyze distortions across countries. Throughout I consider only carbon
emissions stemming from the consumption of fossil fuels as these account for approximately
80% of all anthropogenic carbon dioxide emissions (Schimel et al., 1996).9 More importantly,
this data is widely available due to the tight, physicochemical link between the type of fuel
combusted and the quantity of carbon released (EPA, 2008).10
The total emissions in an economy stemming from fossil fuel combustion can be expressed
by the following identity, Pt tEt, where Pt is carbon emissions, Et is total energy use and trepresents the emissions per unit of energy and can be interpreted as energy impurity. Dividing
both sides of the identity by GDP, relates pollution intensity (emissions per unit of GDP) with
energy intensity (energy per unit of GDP) and impurity:
PtYt
= tEtYt. (1)
Using data on carbon emissions, energy and PPP GDP, I calculate impurity as a residual and
perform the above decomposition in Figure 2(a).11 The figure shows that UK CO2 emission
intensity has followed an inverted-U shape over the 1820-2010 period, whilst energy impurity
largely rose and energy intensity fell. The initial increase in emission intensity was thus driven
by an increase in impurity, whereas the subsequent decline was driven by falling energy intensity.
Next, I examine the likely sources of driving changes in impurity and energy intensity.
Energy Impurity Since carbon emissions are linked directly to the type of fuel consumed, a
changing fuel mix must be the source of rising impurity.12 Figure 2(b) shows how the fuel mix in
the UK has shifted from clean renewable fuels like wood, towards dirty fossil fuels like coal.
According to most international protocols, the burning of biomass materials for energy does not
add to the concentrations of carbon dioxide in the biosphere since it recycles carbon accumulated
by the plant-matter during its lifecycle. The burning of fossil fuels however contributes to higher
8 In the Appendix I demonstrate that the experience of the UK is representative.9 The remaining 20% stems largely from changes in land use - such as deforestation or urbanization.
10 Thus, given energy consumption data, we can accurately infer the quantity of emitted carbon.11 For details on data construction, see the Appendix.12 Suppose there are a number of different energy sources, Ei,t, and each emits a fixed quantity of pollution, i.
Total emissions are given by, Pt
i iEi,t. Dividing both sides by GDP I can write:PtYt
=(
i iEi,tEt
)EtYt,
where Et =
i Ei,t is total energy and the term in brackets is its impurity. As the proportion of energy comingfrom some dirty fuel, ED,t/Et, increases, so does the energy impurity.
-
70.1
1.0
10.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 95 0
= 1
United Kingdom,Intensity and Impurity Indices
Energy Intensity
CO2 Intensity
Energy Impurity
(a) Pollution intensity decomposition, UK.
0%
20%
40%
60%
80%
100%
1800 1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
F ue l
Mi x
, %
United Kingdom ,Fuel Mix
Traditional
Fossil Fuel
Other non-fossil
(b) Fuel mix, UK.
Figure 2: Source of the hump-shaped emission intensity, UK.
levels of carbon concentration since it releases CO2 that had previously been removed or fixed
from the biosphere over millions of years and locked under ground in the form of fossil fuels.13
I argue that the change in fuel mix is driven by the evolution of a countrys economic
structure from agriculture to industry and services. Traditional fuels are relatively abundant in
an agricultural and rural setting and subsistence agriculture lends itself to bio-fuel use. On the
other hand, industrial processes and services require modern types of energy which are more
dependable, have greater flexibility, higher energy density and burn hotter than bio-fuels. Thus
as an economy shifts from an agrarian to an industrialized state, a change in fuel mix will
occur - from renewable biomass materials to (predominantly) fossil fuels such as coal, oil or gas.
This change in fuel mix will in turn contribute to higher energy impurity. The above idea is
relatively well represented in the literature in papers such as Grossman and Krueger (1993) or
Fischer-Kowalski and Haberl, eds (2007).
Energy Intensity I suggest two simultaneous channels that drive falling energy intensity:
differential productivity growth and complementarity. Energy is a special type of good since it
is an input at every stage of production. Consequently, assuming there is technological progress
at each stage of production, energy will likely benefit disproportionately from technological gains
relative to other inputs. If, in addition, energy and non-energy inputs are gross complements in
production, over time an economy will devote relatively more resources to the slower growing
non-energy sector in order to maintain relatively fixed proportions of both types of inputs. This
will contribute to falling energy intensity. At this stage, I do not specify to what extent either
13 Both the Intergovernmental Panel on Climate Change (IPCC) and the US Energy Information Adminis-tration consider biomass emissions to be carbon neutral and recommends that reporters may wish to use anemission factor of zero for wood, wood waste, and other biomass fuels. For details see, Emissions of GreenhouseGases in the United States 2000 (November 2001).
-
8channel operates. Instead, I assume that both channels may be active and pin down their extent
in the calibration, once a structural model has been specified.
3 The Model
In this section I specify a structural model that can capture rising impurity as well as falling
energy intensity and hence can generate the hump-shaped emission intensity documented above.
A calibrated version of the model will then be used to examine sources of cross-country variation
in emission intensity.
Preferences There is an infinitely lived representative agent endowed with a unit of time in
each period. Utility is defined over per-capita consumption of agricultural goods at and non-
agricultural goods ct. To generate structural transformation, I follow Gollin et al. (2002) by
assuming a simple type of Stone-Geary period utility:
U(at, ct) =
a+ u(ct) if at > aat if at a, (2)with lifetime utility being given by:
t=0
tU(at, ct), (3)
where 0 < < 1, is the discount factor and u(ct) > 0. From this setup, we see that once per
capita output in the agricultural sector has reached the level a, all remaining labor moves to the
non-agricultural sector.
Technologies Non-agricultural output (YCt) is produced using labor (LyCt) and a modern
energy inputs (ECt):
YCt =(C(BlCtL
yCt)
1 + (1 C)(BEtECt)
1
) 1
(4)
In the above equation, C is the weight of labor in production, is the elasticity of substitution
between labor and energy, whilst BlCt and BEt are exogenous labor and energy augmenting
productivity terms at time t respectively. Modern energy, is produced using labor (LeCt):
ECt = BeCtLeCt, (5)
where BeCt is exogenous productivity at time t. Since modern energy sector can be considered
to consist of non-agricultural sectors such as the mining, drilling or electricity generation, the
-
9total employment in non-agriculture in the model is given by the sum of employment in both
energy and non-energy sub-sectors, LCt LyCt + LeCt.Agricultural output (YAt) is produced using labor (L
yAt) and a traditional energy input(EAt):
YAt = BAtLyAt
AEAt1A . (6)
In the above equation, A is the weight of labor in production and BAt is productivity at time
t. Unlike in the non-agricultural sector I assume an elasticity of substitution of one between
energy and labor, so that the production function is Cobb-Douglas. This is in line with the
argument put forward by Lucas (2004) that traditional agricultural societies are very like one
another. I take this to mean that the structure of time and labor spent across activities in
traditional agricultural societies remains the same, which would imply a production function
like the above. Traditional energy, is produced using labor (LeAt):
EAt = BeAtLeAt, (7)
where BeAt is exogenous productivity at time t. The traditional energy sector can be taken to
be the gathering of fuel wood or charcoal production etc. and as such total employment in the
agricultural sector is given by the sum of employment in both sub-sectors, LAt LyAt + LeAt.Finally, I assume that the total (and exogenous) size of the labor force is given by, Lt = LAt+LCt.
Notice that I assume traditional energy is only used in agriculture whilst modern energy
is only used in non-agriculture. This assumption is made both for simplicity but also because
long run data on sectoral use of different fuels does not exist. This assumption however is
unlikely to be quantitatively important. Demand for modern energy by agriculture comes from
rich countries where agriculture forms only a small part of the economy. Thus modern energy
will play an important role in agriculture only when agriculture (and hence total demand for
agricultural energy) is relatively small. This is verified later, when I show that the model does
very well in matching cross-country differences in fuel mix.
Pollution Burning a unit of modern energy, ECt, releases PCt units of pollution:
PCt = CtECt (8)
where, Ct, is the coefficient of proportionality. Several comments need to be made about the
above pollution production function. First, due to the assumption that agriculture only uses
traditional energy, only non-agricultural energy will generate emissions. Second, any changes in
modern energy impurity, Ct, will reflect changes in the mix of modern fuels used to generate
energy. In this paper I take changes in Ct as exogenous and attribute them entirely to exoge-
nous technological progress which will result in an (un-modeled) change in the composition of
-
10
modern fuel mix and hence modern energy impurity.14 This turns out not to be an important
assumption for the results of the paper: very similar results will hold even if Ct is assumed to
be constant. This is because - in the data - Ct has remained relatively flat over time. Finally,
notice that I assume that emissions influence neither utility nor productivity - and are thus
simply a by-product of energy consumption and production. This a relatively good assumption
for CO2 - at least historically - since carbon dioxide is a colorless, odorless and tasteless gas and
the concern with climate change is a very recent phenomenon.
Government Finally, I allow government to potentially subsidize/tax modern energy con-
sumption by non-agricultural output producers at a rate t. This policy is supported by lump
sum taxes/transfers on the household.15
Competitive Equilibrium I focus on the (tax-distorted) competitive equilibrium of the
above economy which for every t is defined as the: (1) Price of agricultural and non-agricultural
goods, wage rates, as well as traditional and modern energy prices, {pAt, pCt, wt, peAt, peCt}; (2)household allocations: {at, ct}; (3) firm allocations and emissions {LyAt, LeAt, LyCt, LeCt, PCt}; and(4) energy policy {t, Tt} such that:
(a) Given prices and policy, households allocations maximize equation (2) subject to the budget
constraint of the household: pAtat + pCtct = wt Tt.
(b) Given prices and policy, firms allocations, (3), solve the firms problems in output and
energy sub-sectors for s = A,C: maxLyst,Est pstYstwtLystpestEst and maxLest pestEstwtLest
where peAt = peAt and p
eCt = p
eCt(1 t) and firms emit carbon according to equation (8).
(c) The government period budget constraint is given by LtTt = tpeCtECt.
(d) Goods and labor markets clear: YAt = Ltat, YCt = Ltct. LyAt + L
eAt + L
yCt + L
eCt = Lt.
4 Solution
Quantities The problem is solved in two parts. The first step takes employment across agri-
culture and non-agriculture as given and allocates labor within each sector between energy and
output subsectors. I equate the wage-to-energy price ratios derived from the first order condi-
tions of output and energy firms in each sector. This implies the following distribution of labor
14 Notice that changes in the modern energy fuel mix can encompass both different types of fossil fuels like oil,gas or coal (which generate fixed amounts of carbon emissions per fuel type) but also renewable non-traditionalfuels like wind, solar or nuclear (which, according to the EIA, generate zero carbon emissions).
15 Notice also, that the results remain almost unchanged if instead I chose to place a subsidy on modernenergy producers rather than energy consumers.
-
11
across subsectors within agriculture:
LeAt = (1 A)LAt and LyAt = ALAt. (9)
Due to the Cobb-Douglas structure of agriculture, a constant proportion of agricultural workers
are devoted to the production of energy and non-energy inputs. In non-agriculture however,
since the elasticity of substitution is potentially different from 1, the proportion of workers
devoted to the energy and non-energy sectors is potentially non-constant:
LeCt =
(1
1 + xCt
)LCt and L
yCt =
(xCt
1 + xCt
)LCt, (10)
where, xCt (
C1C
) (BEtBeCtBlCt
)1(1 t). Notice that energy benefits from technological
progress twice - both when it is produced (BeCt) and when it is consumed (BEt) - whereas labor
only benefits once (BlCt). Thus, we may reasonably expect the ratioBEtBeCtBlCt
to increase over
time. If, in addition, the elasticity of substitution between energy and non-energy inputs is less
than one, < 1, these differences in sectoral productivity growth will result in an increase in
xCt over time. Since LyCt/L
eCt = xCt, this will generate a reallocation of inputs from energy
to non-energy production in the non-agricultural sector. Finally, higher subsidies on modern
energy will result in a lower xCt and hence more workers devoted to modern energy production.
The second step determines the division of labor across agriculture and non-agriculture.
Preferences imply that YAt = aLt. Combining this with equations (6), (7) and (9), employment
in agriculture and non-agriculture is respectively given by:
LAt =a
AA(1 A)1ABAtB1AeAtLt and LCt = Lt LAt. (11)
Higher productivity in agriculture results in a smaller proportion of workers being needed to
produce the subsistence level of food and their subsequent reallocation to non-agriculture. Given
the above, equations (5), (7) and (8) determine sectoral energy use and emissions.
Prices Normalizing the wage rate to one, final good firms first order conditions imply that
the price of sector s = A,C goods is pst = 1/YstLst
. Using the above results:
pAt =1
AA (1 A)1ABAtB(1A)eAtand pCt =
1
1C BlCt
(1 + (1t)xCt
) 11
. (12)
Improvements in productivity in agricultural sectors will lead to falling prices of agriculture.
If both labor specific productivity in non-agriculture and xCt increase over time and < 1,
so that the non-agricultural sector substitutes away from modern energy, then prices of non-
agriculture will also fall over time. Taking the first order conditions of the energy firms, the
-
12
price of traditional and modern energy is given by:
peAt =1
BeAtand peCt =
1
BeCt. (13)
Improvements in energy production productivity will result in falling energy prices.
Intensity Next, I examine the evolution of emission intensity over time. Constant price GDP
evaluated at time t prices is Yt = pAtYAt + pCtYCt. Aggregate emission intensity, Nt, can
then be written as the product of the constant price share of non-agriculture in GDP, dt pCtYCt
pAtYAt+pCtYCt, and non-agricultural emission intensity, NCt PCtpCtYCt :
Nt PCtpAtYAt + pCtYCt
= dtNCt. (14)
For illustrative purposes, in this section only, it is helpful to impose the following:
Assumption 1.Bjt+1Bjt
= gj > 1 for j = A, eA, lC,E, eC;gEgeCglC
> 1; t = 0 and Ct = 1.
I then establish the following two theorems describing the evolution of dt and NCt (with
proofs in Appendices 12.3.1 and 12.3.2) to gain insight into the evolution of Nt.
Proposition 1. Suppose Assumption 1 holds. Then when LAt = Lt, dt = 0; when LAt < Lt,
dt+1/dt > 1 and limt dt = 1.
Intuitively, as agricultural productivity rises, less workers are needed to satisfy subsistence.
This results in a reallocation of workers to non-agriculture and an accompanying increase in the
constant prices share of non-agriculture over time.
Proposition 2. Suppose that Assumption 1 holds and < min{
log(gE)log(gE)+log(geC)log(glC) , 1
}then, when LAt < Lt, NCt+1/NCt < 1 and limtNCt = 0.
The intuition here is that if there is enough complementarity between energy and non-
energy inputs, higher technological progress associated with energy will result in a reallocation
of resources away from energy towards non-energy in the non-agricultural sector at a fast enough
pace to engender a monotonic decline in non-agricultural emission intensity towards zero.16
Putting these facts together lends some insight into the evolution of aggregate emission
intensity. When a country is on the brink of industrialization, it is aggregate emission intensity
will be zero, since it is completely dominated by non-polluting agriculture. As the economy shifts
towards dirty non-agriculture, emission intensity will rise at first. However, emission intensity
16 Notice, the condition on elasticity will be tighter than < 1 if and only if geC > glC . If labor productivityin the modern energy sector does not grow too quickly (i.e. geC glC), then < 1 and gEgeCglC > 1 are sufficientconditions for a declining non-agricultural intensity.
-
13
in non-agriculture falls with time, and thus as an ever greater proportion of GDP comes from
non-agriculture, aggregate emission intensity of the economy will also fall.17 Nonetheless, the
exact shape of emission intensity will depend on underlying parameters. Testing the model
thus requires choosing reasonable parameters to see if these can reproduce observed patterns of
energy and emission intensity in the data.
5 Calibration and UK Results
In this section I calibrate the model to the experience of the UK since it is the first country to
industrialize and has excellent long-run data. Furthermore, the UK has some of the smallest
(carbon subsidies) in the OECD (OECD, 2012), making it an excellent choice for the reference
country. The calibrated model will then serve as a baseline that will allow me to derive the
quantitative role that wedges play in determining emission-intensities across countries.
Calibration The model is calibrated to reproduce: 1) traditional and modern energy use;
2) agricultural and non-agricultural employment; 3) total GDP; 4) the (after subsidy, current
price) share of modern energy in non-agricultural value added and 5) the impurity of modern
energy in the UK between 1820 and 2010. In addition, I obtain net fossil-fuel subsidies directly
from UK data. For sources and construction details of all data, see the Appendix.
First, L1990, EC1990 and C1990 are normalized to 1 and Lt, LAt, LCt, ECt, EAt and Ct are
fed into the model directly from the data. Notice that Ct is modern-energy impurity and is
derived as the ratio of carbon-emissions to modern (non-biofuel) energy use.18
Second, I turn to the agricultural sector. I set A = 0.92 to match the proportion of time that
traditional agricultural households devote to gathering fuel wood. Since I lack data for the UK,
I match this parameter to Nepalese data - however the results will be robust to changes in this
parameter. Given the above and using equations (7) and (9), I extract the labor productivity
of the traditional energy sector, period by period:
BeAt =EAt
(1 A)LAt . (15)
Normalizing a to 1 and using equation (11), I extract a measure of agricultural total factor
productivity, period by period:
BAt =a
AA(1 A)1ALAtB1AeAtLt. (16)
17 In fact, if dt and NCt are replaced by their first-order linear approximations, the resulting product will be ahump-shaped, quadratic parabola so that aggregate emission intensity is hump shaped to a first approximation.
18 This parameter refers to modern-fuel impurity, rather than total-energy impurity, t. Modern-fuel impuritycaptures the change in the composition of modern-fuels used to generate energy, rather than the change in thecomposition of total energy from traditional to modern-fuels.
-
14
-200%
-160%
-120%
-80%
-40%
0%
40%
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
Net Subsidy Rates on Carbon Energy, UK
Figure 3: Net Carbon Subsidy Rates in the UK, 1820-2010.
Third, I turn to the non-agricultural sector. Using data from the IFS (2013), the IEA, (BP,
2008) and Mitchell (2011), I construct a measure of taxes on modern fuels in the UK. I then
combine this with data of modern energy subsidies from the OECD (2012) to create a measure
of net subsidies on modern energy in the UK between 1820 and 2010 denoted by, t. The
resulting data is shown in Figure 3. Notice that net subsidies are overwhelmingly negative -
historically the UK has imposed large taxes on carbon energy. This squares well with evidence
from Parry and Small (2005) who argue that the UK has historically taxed fossil fuels heavily.
Using equations (4), (5), (10), (12) and (13), the (after subsidy, current price) share of modern
energy in non-agricultural value added, sect, can be expressed as a function of xCt and t:
sect peCt(1 t)ECtpCtYCt
=1
1 + xCt(1t). (17)
Taking sect and t from the data, I can thus infer that xt =(1sect)(1t)
sect. Consequently, from
equations (5) and (10), I extract the labor productivity of the modern energy sector, period by
period:
BeCt =ECt(1
1+xCt
)LCt
. (18)
Fourth, I choose the elasticity of substitution between labor and energy in non-agriculture to
be = 0.76. This is chosen to lie in the mid-range of the values usually estimated for Allen partial
elasticities between energy and labor in manufacturing. Berndt and Wood (1975) estimate the
elasticity of substitution in US manufacturing between energy and labor to be 0.65. Griffin and
Gregory (1976) estimate this elasticity for numerous advanced European countries and the US
to be between 0.72 and 0.87. Stefanski (2014) estimates the elasticity of substitution between oil
and non-oil inputs in the non-agricultural sector to be between 0.72 and 0.75. Kemfert (1998)
as well as Kemfert and Welsch (2000) estimate this elasticity for Germany to be 0.871. I show
robustness of in section 10.
-
15
Parameter Values Target
BlC1990,BE1990, L1990,EC1990, C1990 1 Normalizationa 1 Normalization
1 A 0.08 Share of time spent gathering fuel wood1 C 0.07 Modern Energy Share in Non-Agr VA, 1990
0.76 Berndt and Wood (1975) and Griffin and Gregory (1976){BAt}2010t=1820 {} Agriculture Empl. Share, t{BlCt}2010t=1820 {} GDP per capita, t{BeAt}2010t=1820 {} Trad. Energy Consumption, t{BeCt}2010t=1820 {} Fossil Energy Consumption, t{BEt}2010t=1820 {} Modern Energy Share in Non-Agr, t{Lt}2010t=1820 {} Labor Force, t{Ct}2010t=1820 {} Emissions/Modern Energy, t{t}2010t=1820 {} IFS (2013), IEA, (BP, 2008), Mitchell (2011)
Table 1: Calibrated parameters
Fifth, I normalize non-agricultural labor productivity and energy specific productivity in
1990 to one, so that: BlC1990 = 1 and BE1990 = 1. This then allows me to infer C from
equation (17) evaluated for 1990 after substituting for xt:
C =1
1 + (1 1990)1 B1
eC1990x 11990
. (19)
Finally, I extract labor- and energy-specific technological progress in the non-agricultural sector.
I feed into the the model - period-by-period - the constant price GDP found in the data so that
GDPDatat = pA1990YAt + pC1990YCt.19 In this expression sectoral outputs can be shown to be
YAt = aLt and YCt = BEtECt(1C)1 (1+ xCt(1t) )1 , whilst 1990 sectoral prices come from
equation (12) and the fact that BlC1990 = BE1990 = 1. Consequently, I extract energy-specific
technological progress in the non-agricultural sector:
BEt =GDPDatat pA1990aLt
pC1990ECt(1 C)1 (1 + xCt(1t) )1
. (20)
Furthermore, since xCt (
C1C
) (BEtBeCtBlCt
)1(1 t), taking xCt from the data, I can
extract labor-specific productivity in the non-agricultural sector:
BlCt =BEtBeCt(
C1C
) 1
(1 t) 1x1
1Ct
. (21)
Table 1 summarizes the calibrated parameters. Figure 4 shows the evolution of implied
productivity parameters and their average growth rates. Growth of TFP in agriculture was
19 Since the wage rate in the model is normalized to one in each period, I ensure that the implied level ofGDP in the model and the data in 1990 match, by normalizing constant price GDP in the data in 1990 by the
level of GDP in the model, GDPmodel1990 w1990(LA1990 +
(11990)+xC19901+xC1990
LC1990
).
-
16
log(y) = 0.017x+2E-15R = 0.99
log(y) = 0.0084x+ 5E-08R = 0.81
0.01
0.10
1.00
10.00
1820 1860 1900 1940 1980
P ro d
u ct i v
i t y, 1
9 90 =
1
BAt
BlCt
(a) TFP in agriculture and laborproductivity in non-agriculture
log(y) = 0.014x + 6E-13R = 0.94
log(y) = 0.0128x + 1E-11R = 0.87
0.01
0.10
1.00
10.00
1820 1860 1900 1940 1980
P ro d
u ct i v
i t y, 1
9 90 =
1
BeCt
BeAt
(b) Labor productivity in energysectors
log(y) = 0.0244x+7E-22R = 0.77
0.01
0.10
1.00
10.00
1820 1860 1900 1940 1980
P ro d
u ct i v
i t y, 1
9 90 =
1
BEt
(c) Energy productivity in non-agriculture
Figure 4: Productivity implied by the model in the UK, 1820-2010.
approximately 1.7% per year over the the entire period. Labor-specific productivity in non-
agriculture grew by 0.84% a year: this stemmed from a productivity growth of approximately
0.5% pre-WWII and a growth rate of 2.3% post-WWII. Labor productivity growth rates implied
in the energy sectors were approximately 1.4% in the traditional energy sector and 1.3% in
the modern energy sector over the entire period. Modern energy-specific productivity grew on
average by 2.4% a year. It was relatively flat until the last decade of the 19th century, after which
it started growing slowly and then picked up significantly in the inter-war period (1919-1939).
This then followed a time of stagnation and collapse in modern energy specific productivity until
1978, after which we saw productivity rising once more.
UK Results Figure 5 shows the baseline simulations for the UK. Increasing productivity in
agriculture results in fewer workers being needed to produce the subsistence level of food and
the subsequent reallocation of workers to non-agriculture (Figure 5(a)). As workers move to the
sector that uses modern-fuels, the share of modern energy consumption rises (Figure 5(b)) and
consequently so does total energy impurity (Figure 5(c)). At the same time, the low elasticity
of substitution between energy and non-energy inputs in the non-agricultural sector, < 1,
coupled with higher energy productivity causes a falling proportion of non-agricultural value
added to be spent on modern energy (Figure 5(h)) and a declining energy intensity (Figure
5(d)). Finally, Figure 5(e) shows the resulting hump-shaped emission intensity curve. Notice
that the difference between the modern energy intensity curve and the pollution intensity curve
in the figure comes entirely from the exogenously changing impurity of modern energy (Figure
5(c)). The two curves follow a similar path, which indicates that the key driver of the hump
shaped emission intensity is the change in the composition of total energy from traditional to
fossil-fuels, rather than the change in the composition of modern energy itself. The remaining
graphs in Figure 5 shows the result for emissions, modern energy use and GDP per worker.
-
17
0%
10%
20%
30%
40%
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
E mp l
o ym
e nt i
n A
g ri c
u lt u
r e, % Agriculture
Employment Share, UK
Model, Data
(a) UK Agricultural Employment Share.
0%
20%
40%
60%
80%
100%
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
Mo d
e rn
E ne r
g y S
h ar e
, %
Share of Modern Energy in Total Energy, UK
Model, Data
(b) UK Modern Energy Share.
0.0
0.5
1.0
1.5
2.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 95 0
= 1
Total and Modern Energy Impurity, UK
Modern ImpurityModel, Data
Total ImpurityModel, Data
(c) Modern and Total Energy Impurity.
0.0
1.0
2.0
3.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 95 0
= 1
Total Energy Intensity, UK
Model, Data
(d) UK Total Energy Intensity.
0.0
0.5
1.0
1.5
2.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 95 0
= 1
Pollution and Modern Energy Intensity, UK
PollutionModel, Data
Modern EnergyModel, Data
(e) UK CO2 Emission Intensity.
0.00
0.50
1.00
1.50
2.00
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 95 0
= 1
PollutionModel, Data
Modern EnergyModel, Data
Pollution and Modern Energy, UK
(f) Total Emissions.
0.0
1.0
2.0
3.0
4.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 95 0
= 1
GDP per Worker, UK
Data, Model
(g) GDP per Capita.
0%
5%
10%
15%
20%
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
Mo d
e rn
E ne r
g y s
h . G
DP ,
%
Share of Modern Energy in Non-Agricultural VA,UK
Model,Data
(h) Share of Modern Energy in Non-Agr VA.
Figure 5: Simulations and data for UK prices, 1820-2010.
-
18
0.0
1.0
2.0
3.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 87 0
= 1Price of Agriculture/Wage,UK
Data
Model
(a) Price of agriculture relative to the wage.
0.0
1.0
2.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 87 0
= 1
Data Model
Price of Non-Agriculture/Wage,UK
(b) Price of non-agriculture relative to thewage.
0.0
0.5
1.0
1.5
2.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 86 0
= 1
Price of Fuelwood/Wage, UK
Data
Model
(c) Price of fuelwood relative to the wage.
0.0
1.0
2.0
3.0
1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
1 87 0
= 1
Price of Coal/Wage, UK
Data
Model
(d) Price of coal relative to the wage rate.
Figure 6: Simulations and data for UK prices, 1820-2010
External Validity Since the parameters of the model are chosen to exactly reproduce the
key quantities in the data, to obtain a test of the mechanism of the model I compare the prices
implied by the model to the corresponding prices found in the data. In section 4, I showed
that model prices depend on the evolution of sectoral exogenous productivity terms. Since I
use the model to extract productivity from the data, if my model is well-specified then the
extracted productivity terms will imply prices that match up well with the data. Figure (6)
shows the evolution of prices of agriculture, non-agriculture, traditional energy and fossil fuel
energy relative to wages in the UK both the model economy and the data.20 The model does
well in matching the decline in prices. This is confirmed in Table 2 which shows the regressions
of the log of relative prices in the data and the model. The coefficients tell us to what extent
a percentage change in the relative prices in the data is associated with the corresponding
percentage change in the model. Perhaps with the exception of the fuelwood prices, where only
limited data is available, the overall fit is very good.
20 For details of data construction and sources see the Appendix.
-
19
Modellog(pa/w) log(pi/w) log(pea/w) log(pei/w)
Data 0.97*** 0.86*** 0.26*** 1.13***(0.03) (0.01) (0.06) (0.01)
Obs. 191 191 50 191R2 0.81 0.98 0.28 0.98
Standard errors in parentheses
*** p
-
20
drive the two components of emission intensity - the constant price share of non-agriculture in
GDP and non-agricultural emission intensity.
Proposition 3. Suppose LAt < Lt, then for given time t: 1)dtDiAt
> 0 and NCtDiAt
= 0; 2)dtDiCt
> 0 and NCtDiCt
< 0; and 3) dt it
< 0 and NCt it
> 0.
Next, I discuss the implications of this theorem using an example. To illustrate the point,
in this section only, I replace all time varying productivity and population growth terms as well
as the subsidy and impurity rates found in the previous section with their 1820-2010 averages.
Agricultural Wedges First, I demonstrate the impact of an agricultural productivity wedge,
DiAt, on emission intensity. In particular, I choose DiAt = 0.29; 0.13; 0.034. This generates
industrializations that begin in 1850 and 1900 and 1980, whilst leaving all other productivity
parameters unchanged. For illustrative purposes, the last agriculture productivity parameter is
chosen to match Chinas employment share in agriculture of 50% in 2000. Figures 7(a)-(c) show
the results for the constant price share of non-agriculture in GDP, non-agricultural emission
intensity and total emission intensity - all computed using 1990 UK prices. From Proposition
3, lower agricultural productivity translates into a lower proportion of non-agriculture in GDP
but leaves the intensity of non-agriculture unaffected. Intuitively, low agricultural productiv-
ity delays industrialization as more workers are needed to satisfy the subsistence requirements.
However, since non-agricultural emission intensity is declining over time - countries that indus-
trialize later will move to a non-agricultural sector that is less energy and emission intensive
than countries that industrialized earlier. This can be interpreted as catching-up to the tech-
nological frontier and means that countries industrializing later, will tend to form an envelope
pattern in emission intensities as demonstrated in Figure 7(c).
Non-agricultural Wedges Not all cross-country differences in productivity however, stem
from inefficiencies in agriculture. In the second counterfactual, I examine the impact on emission
intensity of adding a non-agricultural wedge into the previous counterfactual. I keepDiAt = 0.034
to generate an industrialization that began in 1980 but - in addition - I set DiCt = 0.22, to match
the Chinese-UK ratio of GDP per capita in 2000 of 0.082. Differences in DiAt will now reflect
distortions specific to agriculture - such as soil or climate conditions - whilst differences in DiCt
will capture inefficiencies specific to the non-agricultural sector such as transportation costs,
banking regulations or unionization rates. The results of the second counterfactual are shown
in Figures 7(d)-(f). From the theorem, lower non-agricultural productivity results in a smaller
constant-price non-agricultural share of GDP and a higher non-agricultural pollution intensity.
Intuitively, lower productivity in non-agriculture translates to less non-agricultural output being
produced and hence a smaller proportion of non-agriculture in total output. On the other
-
21
0%
20%
40%
60%
80%
100%
1770 1820 1870 1920 1970 2020 2070
Non-Agr. Share in GDP,(constant prices)
UK
1850start
1900start
1980start
(a) Share of non-Agriculture in GDP,constant 1990 UK prices
0
2
4
6
1770 1820 1870 1920 1970 2020 2070
Non-Agriculture Emission Intensity,relative to UK 1950 Agg. Emission
Intenisty
(b) Non-Agriculture Emission Inten-sity relative to 1950 UK AggregateEmission Intensity
0
0.5
1
1.5
2
1770 1820 1870 1920 1970 2020 2070
Agg. Emission Intensity,relative to 1950 UK
UK1850start 1900
start 1980start
(c) Aggregate Emission Intensity,relative to 1950 UK
0%
20%
40%
60%
80%
100%
1770 1820 1870 1920 1970 2020 2070
C2
Non-Agr. Share in GDP,(constant prices)
(d) Share of non-Agriculture in GDP,constant 1990 UK prices
0
5
10
15
20
25
30
1770 1820 1870 1920 1970 2020 2070
Non-Agriculture Emission Intensity,relative to UK 1950 Agg. Emission
Intenisty
C2
(e) Non-Agriculture Emission Inten-sity relative to 1950 UK AggregateEmission Intensity
0
1
2
3
1770 1820 1870 1920 1970 2020 2070
Agg. Emission Intensity,rel. to 1950 UK
C2
(f) Aggregate Emission Intensity,relative to 1950 UK
0%
20%
40%
60%
80%
100%
1770 1820 1870 1920 1970 2020 2070
Non-Agr. Share in GDP,(constant prices)
C3
(g) Share of non-Agriculture in GDP,constant 1990 UK prices
0
5
10
15
20
25
30
1770 1820 1870 1920 1970 2020 2070
Non-Agriculture Emission Intensity,relative to UK 1950 Agg. Emission
Intenisty
C3
(h) Non-Agriculture Emission Inten-sity relative to 1950 UK AggregateEmission Intensity
0
1
2
3
1770 1820 1870 1920 1970 2020 2070
Agg. Emission Intensity,rel. to 1950 UK
C3
(i) Aggregate Emission Intensity,relative to 1950 UK
Figure 7: Counterfactuals examining the impact of wedges on emission intensity.
-
22
hand, a wedge in non-agricultural productivity results in a composition of the non-agricultural
sector resembling one of the past - when a greater proportion of workers were employed in non-
agricultural energy production and the non-agricultural sector used more energy. This translates
into a higher emission intensity. Since the two effects go in opposite directions, the final impact
on total emission intensity is ambiguous and depends on parameters. In this case, as shown in
Figure 7(f), initially emission intensity is slightly lower as the size effect dominates and then
slightly higher as the intensity effect dominates.
Energy Subsidies The previous wedges - even taken together - cannot capture the large
variation in emission intensity across countries. In the third counterfactual, I choose a modern
energy price wedge to absorb the remaining differences in cross country emission intensities. In
particular, I continue to examine the hypothetical economy with DiAt = 0.034 and DiCt = 0.22
but - in addition - I assume that non-agricultural final good producers also benefit from an
energy price wedge it . I choose it so that the total implicit subsidy is
it =
it + t = 0.64, to
match the ratio between emission intensity in China and the UK in 2000. The results are shown
in Figures 7(g)-(i). As the theorem implies, the subsidy is distortive and lowers non-agricultural
output and hence the share of non-agriculture in GDP. However a higher it also makes the
energy sector more attractive to workers, resulting in higher production of modern energy and
higher emissions. The impact of energy subsidies on emission intensity is again ambiguous.
However, due to the relatively small weight of modern energy in output, the distortion to GDP
is small whilst the impact on emission intensity is large. Hence, as seen in Figure 7(i), aggregate
emission intensity tends to rise with energy subsidies.
The above example demonstrates how the model can be used to disentangle and account
for various distortions within an economy. In the above, I have done this - as an illustrative
example - for China in one particular year. In the next section, I perform the same exercises for
each year and each country in the sample.
7 Results: Model and Data
In this section I use GDP per capita from the Penn World tables, the share of agricultural
employment in the labor force from UNCTAD, and emission intensity data in conjunction with
the calibrated model to infer fossil fuel subsidies for a panel of 170 countries over the 1980-2010
period using the strategy outlined above. I take each country at each point in time to be a closed
economy. I assume each economy is identical in all respects except for the existence of three
wedges (and the size of the labor force). In particular, for each country, i, and in each period,
t, I choose the agricultural productivity wedge, DiAt, to match the ratio between the agricultural
-
23
ALB
ARE
ARG
ARM
AUS
AUT
AZE
BEN
BFA
BGD
BGR
BHR
BHSBIH
BLR
BLZ
BOL
BRA
BRN
BTN
CAF
CAN
CHE CHL
CHN
CIV
CMRCOG
COL
COM
CPV
CRI
CSK
CUB
CYPCZEDEU
DNK
DOM
ECU
EGY
ESP
ETH
FIN
FJI
FRA
GAB
GBR
GHAGMBGNB
GNQ
GRC
GTM
GUY
HND
HRV
HTI
HUN
IDNIND
IRLIRN
IRQISL
ISRITA
JAM
JORJPN KAZ
KEN
KGZ
KHM
KOR
KWT
LAO
LBN
LBR
LBY
LKA
LTULVA
MAR
MDA
MDG
MEX
MKD
MLI
MLT
MNG
MRT
MUS
MWI
MYS
NAM
NER
NGANIC
NLDNOR
NPL
NZLOMN
PAK
PAN
PERPHL
PNG
PRT
PRY
QAT
RUS
RWA
SAU
SENSLB
SLE
SLV
SRBSURSVKSVN
SWESYR
TGO
THA
TJK
TTO
TUNTUR
TZAUGA
UKR
URY
USA
USSRUZB
VNMVUT
WSM
YEM
YUG
ZAF
ZMB
0
.2
.4
.6
.8
1
Trad
ition
al F
uel S
hare
(Mod
el)
0 .2 .4 .6 .8 1Share of Traditional Fuels (Data)
(a) Proportion of traditional fuel impliedby the model versus the data for 2000.
AGO ALB
ARE
ARG
ARM
AUS
AUT
AZE
BEL
BENBGD
BGR
BHR
BIHBLR
BOLBRA
BRN
BWA
CAN
CHECHL
CHN
CIV
CMRCOG
COLCRICUB
CYPCZEDEU
DNK
DOMDZA
ECU
EGY
ESPEST
ETH
FINFRA
GAB
GBR
GEO
GHA
GRC
GTMHND
HRV
HTI
HUN
IDNIND
IRLIRNIRQ ISL
ISRITA
JAMJOR
JPNKAZ
KEN
KGZ
KHM
KOR
KWT
LBN
LBY
LKA
LTU
LUX
LVAMAR
MDA
MEXMKDMLT
MNG
MYS
NAMNGA NIC
NLDNOR
NPL
NZLOMN
PAK
PAN
PERPHL
POLPRT
PRY
QAT
RUS
SAU
SEN
SGP
SLV
SRBSVKSVN
SWESYR
TGO
THA
TJK
TKM
TTO
TUNTUR
TZA
UKR
URY
USA
UZBVEN
VNM
YEM
ZAF
ZMB
.01
.1
1
10
Mod
ern
Ener
gy re
l. to
UK
(Mod
el)
.01 .1 1 10Modern Energy rel. to UK (Data)
(b) Modern Energy relative to the UK inModel and data for 2000 (Log Scale).
Figure 8: Model Fit
employment share in the particular country and that of the UK, I choose the non-agricultural
productivity wedge, DiCt, to match the ratio between the specific country and the UK, and I
choose a fossil fuel subsidy wedge, it , to match the ratio between emission intensity of the
country and the UK. I thus force the model to reproduce cross-country differences in GDP-per-
worker, agricultural employment shares and carbon emission-intensities and hence also relative
carbon emissions-per-worker.
Traditional and Modern Energy The model does well in matching moments of the data to
which it has not been calibrated. Figure 8(a) plots the share of traditional fuels in total energy
consumption from the model versus the corresponding number from the data for the year 2000.
Traditional Fuel data comes from Krausmann et al. (2008b) who perform a very careful global
energy accounting exercise for 175 countries in the year 2000. Importantly, the traditional fuel
measure is comparable to the data used to calibrate the UK model. We see that the model does
extremely well in predicting the share of energy from traditional sources across countries. Figure
8(b) plots the modern energy relative to the United Kingdom consumed by each country for the
model and the data in 2000. The model also does extremely well. The good fit of the model
along this dimension is a consequence of the fact that modern energy use is vastly dominated
by carbon energy, which in turn is closely tied to carbon emissions which we have calibrated the
model to match. In the above exercise I show only the year 2000, since this is the only year for
which (comparable) traditional fuel data exists. In the appendix, I use a panel of traditional
fuel data from the WDI (2013) - which whilst not exactly comparable to the previous data -
nonetheless shows a similar relationship holds over time.
Prices Next, I compare the predictions of the model with respect to prices. In the Appendix,
I construct a panel of agriculture and non-agriculture price-to-wage indices for each country
-
24
Modellog(pa/w) log(pc/w) log(pa/w) log(pc/w)
Data 0.48*** 1.13*** 0.63*** 0.96***(0.01) (0.02) (0.01) (0.01)
Obs. 3353 3353 899 899R2 0.40 0.72 0.70 0.91Country FE yes yes yes yesSample All All OECD OECD
Standard errors in parentheses
*** p
-
25
log(1 iModel,t)
log(1 iData,t) 0.68*** 0.62*** 0.64*** 0.98*** 0.79*** 1.06***(0.04) (0.04) (0.04) (0.08) (0.09) (0.06)
Obs. 1,057 1,057 1,057 257 257 257R2 0.20 0.23 0.89 0.38 0.45 0.89Time FE no yes no no yes noCountry FE no no yes no no yesSample All All All OECD OECD OECD
Standard errors in parentheses*** p
-
26
interpretation. Since the model-based wedges provide a broader measure of support than subsi-
dies extracted using the price-gap method - the deviation between the two can be interpreted as
capturing all indirect distortions not reflected in at-the-pump gasoline prices. Thus, for example,
we can infer that countries below the 45-degree line and in the bottom left quadrant of Figure 9
(such as China or Russia), provide more support to fossil fuels than is captured in the gasoline
price. Examples of such policies could be the guaranteeing of bank loans to energy intensive
industries, providing disproportionately generous benefits to coal miners or protecting energy
intensive industries from outside competition. Countries above the 45 degree line and in the top
right quadrant - like France - are those that in addition to high taxes also have other policies
to discourage fossil fuel consumption. For example, in France, there are strong restrictions on
generating energy from fossil fuels and an emphasis on nuclear energy. Whilst these types of
policies are not a direct tax on fossil fuels, they nonetheless discourage the use of fossil fuels and
hence shows up in our measure of the energy wedge. Whilst this is a useful method that helps
to interpret some of the data, it also has to be used with caution. Since wedges are catch-alls,
they include all policy and non-policy deviations that encourage or discourage countries to use
disproportionately more or less fossil fuels. Thus, undoubtedly, some deviation between model
and data stems from non-policy factors such as endowments or geographical location. Nonethe-
less, since the model does well whilst controlling for country-fixed effects, these considerations
are probably secondary.
8 Examining Implicit Wedges
Figure 10(a) presents the subsidy rates, the (current price) share of subsidies in GDP as well
as total and per worker subsidies (in 1990 international dollars) for the largest 20 subsidizers
in 2010. Figure 10(b) shows a map of subsidy rates around the world. Oil rich countries as
well as current/ex-communist countries have the highest subsidy rates. These countries tend to
have a history of direct subsidies on dirty energy and supporting heavy industries that use large
quantities of energy. In the framework of the model, these policies are captured as subsidies on
carbon energy. In absolute terms, China, the US, Russia and India are the largest subsidizers.
The carbon wedges in these countries cost approximately 717 billion dollars (in international
1990 dollars) in 2010 alone. Countries in Europe tend to have implicit taxes on carbon energy.
This captures either high levels of direct energy taxation or - for instance - the fact that many
European countries subsidize green energy or discourage carbon energy through high direct or
indirect taxation - which is effectively captured in the model by a tax on carbon energy. Finally,
countries in sub-saharan Africa as well as South America have the highest implicit taxes. This
may be due to high taxes on energy, tariffs on imported fossil fuels, the closed nature of those
-
27
Rank Subs Rate Subs/GDP Subs. Total Subs. per cap.(%) curr. prices, (%) bill. 1990 int. $ 1990 int. $
1 KWT 88.56% KWT 34.41% CHN 419.8 QAT 65092 TTO 87.08% TTO 27.13% USA 169.7 TTO 47313 SAU 84.40% SAU 25.40% USSR 122.5 KWT 45254 QAT 84.25% ZAF 25.19% RUS 83.6 SAU 30125 OMN 84.15% KAZ 24.43% IND 43.9 OMN 25556 KAZ 83.74% UZB 22.28% IRN 29 BRN 22517 ZAF 81.68% QAT 21.83% SAU 27 ARE 21758 UZB 79.71% BIH 21.34% ZAF 25.9 BHR 19399 BHR 77.40% BHR 19.63% UKR 14.5 KAZ 172310 BIH 75.76% OMN 18.05% KAZ 14.3 ZAF 145011 EST 74.49% EST 16.08% KOR 13.7 EST 141612 IRN 73.68% VEN 15.57% CAN 12.8 TKM 116613 CHN 73.58% UKR 14.75% AUS 11.8 LBY 115114 GNQ 70.21% MNG 14.09% VEN 9.7 RUS 112615 RUS 69.76% LBY 13.71% POL 9.4 IRN 110416 VEN 69.51% RUS 13.56% MYS 9.1 USA 108917 BRN 68.99% IRQ 13.30% THA 8.2 AUS 102918 UKR 68.96% USSR 12.59% ARE 7.1 BIH 89319 LBY 68.51% IRN 12.27% UZB 5.7 USSR 85720 MNG 68.40% BRN 12.05% KWT 5.7 VEN 763
(a) Top 20
(.75,1](.5,.75](.1,.5](.1,.1](.5,.1](.75,.5][1000,.75]No data
(b) Map of subsidy rates in model, 2010.
Figure 10: Implied subsidies around the world, 2010.
-
28
Weighted Mean
1
.75
.5
.25
0
.25
.5
Subs
idy
Rate
s (%
)
1980 1990 2000 2010
(a) Average global fossil fuel subsidy rates.
Total Subsidies, World(Billions, 1990 PPP US $)
200
400
600
800
1000
Subs
idie
s
1980 1990 2000 2010
(b) Total global fossil fuel subsidies.
Figure 11: Evolution of Global Subsidies.
economies or their distance from energy producers, which - from the perspective of the model -
is reflected as implicit taxes on carbon energy.
Next, I examine the rate and level of fossil-fuel subsidy wedges over time. Figure 11(a) shows
the global (emission-weighted) mean of implied subsidy rates over time. Globally, subsidy rates
are roughly constant until the late 90s after which there is a large increase. Figure 11(b) confirms
the trend by showing the global sum of inferred subsidies over the period. Until the end of the
90s subsidies are roughly around 200 billion dollars a year. After that there is a massive increase
and by the end of 2010, subsidies have reached a staggering 983 billion US dollars a year. Figure
12 then presents the corresponding graphs for some interesting countries. China, the USA,
the ex-USSR and India are the largest subsidizers in the world. Observe that each one of the
countries has seen a large increase in implicit fossil-fuel wedges, but none more so than China,
where implicit subsidies increased from roughly 100 billion dollars in the late 90s to over 400
billion dollars by 2010. As can be seen from Figure 12(a), this was largely driven by an increase
in indirect subsidies, not captured by the price gap method but detected by the implicit wedges
of the model. Also, included in the figure are graphs for Saudi Arabia - a country with one of the
highest rates of subsidization in the world - and for Germany a country with very high implicit
taxes on fossil fuel energy. In both countries there is an increase in subsidies or a decrease in
taxes, after the late 90s. Notice, that most of the distortions are direct since there is agreement
between the implicit wedges and the subsidies extracted using the price-gap approach.
9 Counterfactuals
I use the model to examine the impact that the above wedges have on global emissions and GDP.
In particular, I assume that the evolution of productivity in each country stays the same, but I
set all positive energy subsidies to zero in all countries and calculate the resulting counterfactual
-
29
China
1.5
1
.5
0
.5
1
Subs
idy
Rate
(%)
1980 1990 2000 2010Subsidy (Model) Subsidy (Data)
(a) Sub. Rate: China
USA
1.5
1
.5
0
.5
1
Subs
idy
Rate
(%)
1980 1990 2000 2010Model Data
(b) Sub. Rate: USA
USSR
1.5
1
.5
0
.5
1
Subs
idy
Rate
(%)
1980 1990 2000 2010Subsidy (Model) Subsidy (Data)
(c) Sub. Rate: USSR
China
200
100
0
100
200
300
400
Subs
idie
s, B
illio
ns 1
990
PPP
US$
1980 1990 2000 2010
(d) Sub. Quantity: China
USA
200
100
0
100
200
300
400
Subs
idie
s, B
illio
ns 1
990
PPP
US$
1980 1990 2000 2010
(e) Sub. Quantity: USA
USSR
200
100
0
100
200
300
400
Subs
idie
s, B
illio
ns 1
990
PPP
US$
1980 1990 2000 2010
(f) Sub. Quantity: USSR
India
1.5
1
.5
0
.5
1
Subs
idy
Rate
(%)
1980 1990 2000 2010Subsidy (Model) Subsidy (Data)
(g) Sub. Rate: India
Saudi Arabia
1.5
1
.5
0
.5
1
Subs
idy
Rate
(%)
1980 1990 2000 2010Subsidy (Model) Subsidy (Data)
(h) Sub. Rate: S. Arabia
Germany
1.5
1
.5
0
.5
1
Subs
idy
Rate
(%)
1980 1990 2000 2010Subsidy (Model) Subsidy (Data)
(i) Sub. Rate: Germany
India
100
75
50
25
0
25
50
Subs
idie
s, B
illio
ns 1
990
PPP
US$
1980 1990 2000 2010
(j) Sub. Quantity: India
Saudi Arabia
100
75
50
25
0
25
50
Subs
idie
s, B
illio
ns 1
990
PPP
US$
1980 1990 2000 2010
(k) Sub. Quantity: S. Arabia
Germany
100
75
50
25
0
25
50
Subs
idie
s, B
illio
ns 1
990
PPP
US$
1980 1990 2000 2010
(l) Sub. Quantity: Germany
Figure 12: Subsidy rates and quantities in model and data (rates only).
-
30
Observed Emissions
Counterfactual Emissions4
5
6
7
8
9
Met
ric T
ons
of C
arbo
n (b
illion
s)
1980 1990 2000 2010
(a) Emissions with and without subsidies
Relative GDP, World(No Subsidies/Subsidies)
1
1.005
1.01
1.015
GDP
Dis
torti
on (%
)
1980 1990 2000 2010
(b) Relative GDP with and without sub-sidies
Figure 13: Effect of Subsidies
estimate of global emissions and GDP.
First, the impact of this experiment on emissions is shown in Figure 13(a). In 2010 alone,
annual emissions would have be 36% lower were it not for massive fossil fuel wedges. Over the
1980-2010 period, cumulative emissions would have been 20.7% percent lower if countries had
not subsidized fossil fuels. Removing these wedges can thus massively lower carbon emissions
without the painful costs associated with other schemes designed to lower carbon emissions.
Second, the elimination of subsidy distortions would have resulted in higher output over the
period. Figure 13(b) shows the ratio of GDP in this counterfactual world relative to observed
GDP. Prior to the late 90s annual GDP would have been higher by approximately 0.3% on a
global scale. The increase in output associated with eliminating wedges rises to approximately
1.7% of global GDP or approximately 838 billion US dollars by 2010. Importantly, this is in
addition to the direct cost of the wedges. Notice that the effects are even larger for particular
countries. For example, Chinas GDP would have been higher by 6.1% or approximately 431
billion US dollars in 2010 alone. These are massive effects. Finally, the vast bulk of these wedges
(73% in 2010) came from only four countries: China, the US, India and Russia.
10 Robustness
Elasticity I now examine the robustness of the model to changes in the energy-labor elasticity
of substitution, . In particular, I vary the elasticity between 0.1 (near-Leontief) and 10 (near-
perfect substitution), re-calibrating the model each time. Table 5 shows the results. Column
(1) presents the implied average growth rate of energy-specific technological progress between
1820-2010 for each choice of .22 Choosing > 1, results in growth rates less than or equal to
zero. Over such a long period of time, a negative or even zero technological growth rate seems
22 These are derived as the slope of the regression of the log of the implied values of BEt on time.
-
31
(1) (2) (3) (4) (5) (6) (7) (8)Model to Data Subs. Subs.2010 Subs.2010 CO2 GDP
gE Elast., OECD Total /Subs. 1980 /Ave. Subs /CO2 /GDPpc,t Subs. 2010 1980-1999 2010 2010
0.10 0.7% 14.44 3.53 1598 2.2 2.3 0.67 1.0360.30 0.9% 1.15 1.57 1380 2.3 2.4 0.67 1.0270.60 1.5% 0.96 1.09 1088 2.9 3.2 0.65 1.0190.76 2.4% 0.94 0.98 983 3.4 3.7 0.64 1.0170.90 5.7% 0.93 0.92 911 3.7 3.8 0.63 1.016
1.10 -5.6% 0.92 0.87 832 4.3 3.7 0.62 1.0151.30 -1.8% 0.91 0.83 769 4.2 3.5 0.61 1.0151.50 -1.0% 0.91 0.80 721 4.0 3.3 0.60 1.0155.00 -0.1% 0.89 0.66 591 3.8 2.8 0.44 1.02210.00 0.0% 0.89 0.63 561 3.7 2.6 0.40 1.025
Table 5: Effects of varying on the results. For details, see text.
highly implausible. Realistically, the parameter should thus be smaller than one.23
Choosing too small a , however, also contributes to unrealistic results. Column (2) shows
the slope parameters from the regression of the log of non-agricultural price to wage ratio in the
model to that of the data in the OECD (controlling for time fixed effects), whilst column (3)
shows the slope of the regression of log(1 it ) in the model to that of the data in the OECD.24Choosing < 0.6, results in prices and subsidies in the model that change more than those
in the data. A realistic estimate of would thus lie between approximately 0.6 and 0.9. This
squares remarkably well with the empirical evidence discussed in section 5, which points towards
being between 0.65 and 0.871.
Column (4) of the table shows implied global subsidies in 2010. As increases, implicit
subsidies decrease. In the realistic range of , subsidies lie between 911 billion and 1.1 trillion
dollars. The predicted subsidies are also large for other values of . Furthermore, column
(5) - which shows the ratio of global subsidies in 2010 and 1980 - demonstrates that subsidies
are growing over time for all values of . Next, column (6) shows the ratio of implicit global
subsidies in 2010 with respect to average, annual global subsidies between 1980 and 1999. For
values of < 1.5, columns (5) and (6) are very similar, suggesting that the finding that most of
the increase in subsidies occurred post-1999, is robust to other values of . Finally, columns (7)
and (8) show the ratio of counterfactual emissions and GDP to observed emissions and GDP
respectively in 2010, where the counterfactual sets positive subsidies to zero. A higher implies
a greater positive impact of removing subsidies on GDP and emissions. For reasonable values
23 Recall that when = 1, the production function collapses to a Cobb-Douglas form and energy specifictechnological progress cannot be identified.
24 For = 0.76, these are simply the coefficients found in columns 4 of Tables 2 and 4.
-
32
of emissions would be 1/3 lower without subsidies whilst GDP would be 1.6-1.9% higher.
The above exercise demonstrates that the key results of the paper are both qualitatively and
quantitatively robust to different choices of .
Trade In the baseline model I had assumed that all energy is produced locally. As I now show,
this model is isomorphic to one where all fossil-fuels are imported from abroad. Suppose that
in the place of the domestic fossil-fuel energy sector, there is instead an export sector producing
a good that can be traded for imported fossil-fuels with the following production function:
XCt = BxtLxCt. (22)
This good can then be exchanged for fossil-fuel energy from abroad according to the following
balanced budget condition:
XCt = pEt (1 +
Tt )ECt, (23)
where pEt is the exogenous international price of fossil fuels and (1 + Tt ) captures a country-
specific transportation cost. Substituting equation (23) into (22) I obtain:
ECt =Bxt
pEt (1 + Tt )
LxCt. (24)
Thus, setting BxtpEt (1+
Tt )
= BeCt, this condition collapses to equation (5). The baseline model
and the model with energy imports are thus quantitatively and qualitatively identical.25
11 Conclusion
Countries exhibit emission intensities that are hump-shaped with income. This paper demon-
strates that industrialization drives this pattern whilst deviations from it are symptomatic of
wedges to sectoral productivity or to the price of fossil-fuels. I use a calibrated model of struc-
tural transformation to disentangle these distortions and to measure subsidy-like wedges on
fossil-fuels for 170 countries between 1980 and 2010. The data is comparable across countries
and time and - unlike measures based on the price-gap approach - it also captures indirect sup-
port to fossil-fuels and controls for price differences across countries. Since this is also the most
comprehensive database of its kind, it should prove to be a crucial tool for both policy makers
and researchers.
25 It is also easy to generalize the baseline model to simultaneously have local and imported fossil-fuel energy.In addition to the above conditions, I can assume that a country has an exogenous endowment of naturalresources, Dt, which allows it to produce fossil-fuels locally: EHCt = BHt(L
hCt)
(Dt)1 . These are perfectsubstitutes to imported fossil fuels. Choosing as a labor elasticity of energy production from internationaldata, I can set BHt(Dt)
1 to match a countrys import share of foreign fossil-fuels. Using this modified modelto infer international energy wedges, I find results to be quantitatively and qualitatively similar to the baseline.These results are available on request.
-
33
Finally, since my method of estimating wedges is model-based, it allows me to perform
counterfactuals and hence to measure the impact of those wedges on emissions and growth. I
find that subsidy-like wedges are a massive contributor to global carbon emissions, accounting for
more than a quarter of emissions over the last thirty years. The direct cost of these distortions
was 983 billion dollars in 2010 alone. Furthermore, by distorting energy prices, subsidy-like
wedges also resulted in global GDP that was 838 billion dollars lower in 2010 than what it
would have been without those wedges. Together, the cost of these distortions amounted to a
staggering 3.8% of total world GDP in 2010. To put this into the starkest possible perspective,
the 2014 IPCC report estimates that climate change will lower global GDP by at most 2% in
50 years. By this measure, subsidy-like wedges on fossil fuels are potentially more damaging
than climate change. Worryingly, using the long time-series data arising from my method, I
also find that wedges are increasing over time. Whilst not all of the estimated distortions can
be eliminated, removing even some of these subsidy-like wedges can potentially help strained
government budgets, make a significant (and importantly cheap) contribution to the fight against
climate change and result in higher levels of global GDP.
-
34
For Online Publication
12 Appendix
hrs/person/day hrs/HH/day
Activity Men Women Children Household HH Shares
Field Work 3.10 2.75 0.05 6.00 27.33%Grazing 0.00 0.00 1.80 5.40 24.60%Cooking 0.38 2.10 2.48 11.28%Grass Collection 0.35 0.98 0.28 2.15 9.79%Water 0.10 1.15 0.23 1.93 8.77%Fuel Wood Collection 0.13 1.15 0.13 1.65 7.52%Employment 0.80 0.13 0.93 4.21%Food processing 0.20 0.70 0.90 4.10%Leaf Fodder 0.10 0.35 0.03 0.53 2.39%
TOTAL 5.15 9.30 2.50 21.95 100.00%
Source: Kumar and Hotchkiss (1988), Table 5.Data constructed by assuming five people/household.
Table 6: Patterns of time allocation in Nepal for Men, Women, Children and Households.
12.1 Data Sources
Energy Share, Agriculture Table 6 shows time allocation information for men, women and
children for the year 1982 in Nepal. The table is constructed from numbers reported by Kumar
and Hotchkiss (1988) and is based on data collected by the Nepalese Agriculture Projects Service
Center; the Food and Agricultural Organization of the United Nations and the International
Food Policy Research Institute.26 In particular, the first three columns of the table show
the number of hours per person per day devoted to a particular activity. Kumar and Hotchkiss
(1988) present the data disaggregated by season - the data in the above table, is aggregated by
taking inter-seasonal averages and hence represents an annual average. To see what fraction of
total hours worked in agriculture is devoted to fuel collection, I construct hours spent per activity
for a typical Nepalese household/agricultural producer. According to the Nepalese Central
Bureau of Statistics,27 the average size of an agricultural household in Nepal is approximately
5 people - a man, a woman and three children. Thus, to obtain the total hours devoted to each
26 Nepal Energy and Nutrition Survey, 1982/83, Western Region, Nepal.27 http://www.cbs.gov.np/nlfs %20report demographic characteristics.php.
-
35
activity for an average household, the men, women and children columns are summed with the
childrens column weighed by a factor of three. This gives total hours per day spent by a typical
Nepalese agricultural household/producer in each one of the above activities. From this, the
fraction of time spent on fuel wood collection is approximately eight percent, which implies that,
1 A = 0.08.
United Kingdom Prices Wage data for craftsmen and farmers (1820-1914) comes from Greg
Clark, on the Global Price and Income History Group web site.28 The wage data for craftsmen
is extended to 2010 using Officer (2011). The data for fuel wood prices also comes from Greg
Clark in the same data file for the year 1820-1869. For the price of fossil fuels, I take the price
of coal in constant 2000 GBP from Fouquet (2011). The assumption here is that the price of
energy is equalized across fossil fuel types. I convert this price back into nominal GBP using
the CPI from Officer (2011).
To construct the price indices for agriculture and non-agriculture, I first build constant and
current price measures of agricultural and non-agricultural value added in the UK. To construct
constant price sectoral value added, I take data from the UN for 1970-2010 on one digit ISIC
v.3 sectoral value-added data in constant 1990 local currency prices. I extend this back to 1855,
using sectoral value added growth rates from the Groningen Growth and Development Center
(GGDC). In particular I use the 10-Sector Productivity Database by (Timmer and de Vries,
2007) and the Historical National Accounts by Smits et al. (2009). Finally, I extend the resulting
sequence of data back to 1820 using growth rates from Broadberry and van Leeuwen (2010).
Second, to construct current price sectoral GDP, I first obtain nominal aggreg